Comment by jaeyounkg | Hacker News Reader

Comment by jaeyounkg | Hacker News Reader

This was an fun read, as someone who's both a Korean BW player and a speech recognition researcher.

It's interesting to note that the original Korean transcription already has many errors, seemingly (and impressively) corrected by LLMs later on. For example, 12 안마당 빌드 (12 courtyard build) is actually 12 앞마당 빌드 (12 frontyard build), which might have been more understandable to BW players. Similarly 투에처리 빌드 (processing-at-two build? makes no sense lol) should have been transcribed 투해처리 빌드 (two-Hatchery build).

Therefore it may also be helpful to directly feed the slang dictionary into Whisper's inference process using contextual biasing. There are lots of ways to do this, but the simplest would be to increase the probability of slang words in the dictionary in the final prediction layer of Whisper by a constant factor. This is fairly easy to implement, for example by using HuggingFace's library: https://huggingface.co/docs/transformers/en/internal/generat...

I am a StarCraft fan and I have no idea what a courtyard or a frontyard is supposed to be! However I do know that the names of buildings, units, technologies, and strategies are usually heavily abbreviated in English. Perhaps the same is true in Korean? A 12 barracks build would usually just be called "12 rax", a two hatchery mutalisk build would be called "2 hatch muta", and a three hatchery hydralisk timing attack / all-in would be called "3 hatch hydra bust".

I believe the equivalent term used in English (exhibited in the new translation) is "natural", short for "natural expansion", which refers to the obvious location where the player should build their first expansion. It sounds like the term used in Korean for this concept literally means "front yard" rather than matching the English term.

Makes sense. And presumably the 12 means that you expand to your natural ("courtyard") with your 12th worker unit (probe, in the case of protoss).

Not the parent commenter but not always. 9 pool just means you build a spawning pool at your main, for instance. This worker-prefix building build-order naming system also breaks down once people start referencing builds like 2 rax academy, 3 hatch muta, etc.

Right, "9 pool" means build a spawning pool when you have 9 workers. So "12 courtyard" means build an expansion when you have 12 workers.

I think strictly "9 pool" means you build the pool when you have 9 supply. However, before you build a spawning pool, the only thing you can build that consumes supply is workers.

A lot of Korean slang is a little different. Source: not Korean but have been in the English community a long time and picked some stuff up.

"1rax double" is equivalent to "1rax expand" or "1rax CC". They use multi or double to mean expand in the early game. Instead of "cheese" or "all-in" they use "pil-sal-gi" which means ace/joker card or "han-bang" which means an army or attack on few resources.

I am not sure what short-hand they use for barracks, gateway, etc.

Instead of "cheese" or "all-in" they use "pil-sal-gi" which means ace/joker card

That’s a really interesting one to me! One thing I’ve noticed is that Koreans do not seem to have the same hangups / negative attitude towards cheese strategies as westerners do!

As far as I can tell, there's no hang up about 'cheese' at the higher levels of competition even among westerners. But that might just be from the extreme Korean influence at that level?

The attitude seems to be that throwing in the occasional cheese is not so much meant to win the game, as it is meant to make sure your opponent wastes resources on defending against a potential cheese at all the other times.

This is very similar to the function of a bluff in a theoretical analysis of poker. Very simplified, the optimal frequency of bluffing is when bluffing just about breaks even against optimal play from your opponents. But throwing the bluffing in masks when you actually have good cards.

A failed cheese usually leaves the attacker so weak that the game is already lost. Cheeses are intended to win the game.

But if you are known to never cheese, your opponent might bet on greedy strategies, sometimes known as "economic cheese": you don't prepare any defense, and skip scouting, to build an overwhelming army all of a sudden at some given time like just after an important couple of upgrades that boost the army (a timing attack).

The "normal" play (economic growth plus scouting) is usually the superior strat, but if your scouting fails to detect a cheese attempt that must be countered with a very specific defense, the game is lost. The occasional cheese keep the players honest so they spend resources in scouting, instead of going greedy.

StarCraft has its own bluffing scheme, that is faking a build so the opponent goes for a specific counter, but actually going for something else.

This is a good example of a game-theoretic equilibrium :)

> A failed cheese usually leaves the attacker so weak that the game is already lost. Cheeses are intended to win the game.

> But if you are known to never cheese, your opponent might bet on greedy strategies, sometimes known as "economic cheese": you don't prepare any defense, and skip scouting, to build an overwhelming army all of a sudden at some given time like just after an important couple of upgrades that boost the army (a timing attack).

Yes, that's exactly what I was trying to say.

And the optimal cheese frequency is when cheesing has the same expected win-rate as 'normal' play.

> StarCraft has its own bluffing scheme, that is faking a build so the opponent goes for a specific counter, but actually going for something else.

Yes. I didn't say cheese was bluffing. Just that the strategic considerations around cheese frequency are similar to the math for bluffing in poker.

Westerners do use cheese but many of them are very unhappy about it. Probably its biggest detractor is Dan 'Artosis' Stemkosky, a man who has dedicated his life to StarCraft and who sees cheese strategies as a betrayal of the beauty of the game. He nevertheless grudgingly engages in the occasional cheese, though his opponents nearly always see it coming because he saves it for when he's truly on tilt (another poker term).

I think the real issue is that, like learning to play the piano, StarCraft demands extreme levels of practice to master its physically demanding control scheme. To then lose to an inferior opponent who merely bluffed you feels profoundly unfair. For whatever reason, Koreans seem to be better equipped to handle the cognitive dissonance associated with such an unfair system. Perhaps the Korean school system (and its infamous final exam) has something to do with it?

Is there any link to “ace” meaning a tennis serve that the defending player fails to make any contact with? I could see the parallel with a “cheese” strategy being an unexpectedly fast attack.

I suspect the link is with playing cards and gambling?

Thanks for the added context on the builds! As "foreign" BW player and fellow speech processing researcher, I agree shallow contextual biasing should help. While not difficult to implement, most generally available ASR solutions don't make it easy to use. There's a PR in ctranslate2 implementing the same feature so that it could be exposed in faster-whisper: https://github.com/OpenNMT/CTranslate2/pull/1789

Do they actually use the Korean word for, like, tossing something to refer to the Protoss? That’s a pretty funny cross-language pun if so.

Haha, no I acutually never associated this with the English word toss lol.

Half of the words in the Korean blurb are just romanizations. Even build is just bil-deu

No, Protoss is just 토스, which is just hangulization of "Toss" aka Protoss.

Been a while since I played the game, but in all translations I was confused why you'd want to build 12 spawning pools.

If I remember correctly, that building was for enabling creation of zerglings and other units in hatcheries (and also for researching upgrades) - but one building was enough to unlock those units in all hatcheries; and it did not produce any units itself, so building more of them wouldn't increase your unit output either.

You could in theory built several of them to research multiple upgrades in parallel, but there were only like 3 possible upgrades anyway, so it would make sense to build 12 of them.

The only reason I could think of would be as a sort of redundancy, so you can keep building zerglings even if the enemy destroys some of the pools. But 12 also seems excessive for that.

So what exactly was the motivation here?

Sorry if I'm talking rubbish here, as I said, it's been a while.

The number is not the amount of buildings, but the amount of drones you have when you build the building

OK, that makes more sense. Thanks a lot!

12 Pool means build your spawning pool at 12 supply, usually the number before the building is at which supply you build the building, and it assumes you're constantly producing workers.

Ooh ok. I feel stupid now :)

Thanks for the clarification!

Reply on news.ycombinator.com